Discrete element contact parameter prediction method and system based on feature analysis-machine learning

Abstract

The invention discloses a discrete element contact parameter prediction method based on feature analysis-machine learning, which comprises the following steps: selecting a discrete element linear parallel bonding model for contact parameter calibration according to soft rock particle characteristics and mold material parameters, and determining model parameters and a value range; obtaining contact parameters of the discrete element linear parallel bonding model based on a small amount of uniaxial compression and a Brazilian splitting test, and forming a contact parameter-soft rock macroscopic feature low-fidelity data set based on the contact parameters; key contact parameters influencing the macroscopic features of the soft rock are determined based on grey correlation analysis GRA, Pearson correlation analysis and mutual information MI; obtaining a contact parameter-macroscopic feature high-fidelity data set based on a large number of uniaxial compression and Brazilian splitting tests; establishing a discrete element linear parallel bonding model key contact parameter prediction model based on machine learning; and determining an optimal prediction model of the key contact parameters of the linear parallel bonding model. The invention further discloses a corresponding system, electronic equipment and a computer readable storage medium.

Description

Technical Field

[0001] The present invention relates to the technical field of rock mechanics and materials science detection, and particularly relates to a discrete element contact parameter prediction method and system based on feature analysis - machine learning. Background Art

[0002] The main problems existing in the current discrete element contact parameter prediction are as follows:

[0003] (1) Low efficiency, long time consumption, and large computational amount: Traditional discrete element contact parameter calibration methods usually require a large number of experiments and numerical simulations, resulting in low efficiency and long time consumption. Especially in complex rock mechanics problems, the calibration process of contact parameters often requires repeated adjustment and verification, with a huge computational amount.

[0004] (2) It is difficult to select the optimal model with a single - level performance index: In the traditional parameter calibration process, usually relying on a single mechanical performance index to evaluate the accuracy of the model, this may lead to poor performance of the model in other performance indices and it is difficult to select the optimal model.

[0005] (3) Lack of an efficient parameter prediction method: Most of the existing discrete element contact parameter prediction methods rely on empirical formulas or manual adjustment, lacking systematicness and automation, and are difficult to handle complex rock mechanics problems. Summary of the Invention

[0006] In view of this, to solve the problems in the background art, the present invention proposes a discrete element contact parameter prediction method and system based on feature analysis - machine learning. By combining methods such as grey relational analysis (GRA), Pearson correlation analysis, and mutual information (MI), the key contact parameters affecting the macroscopic mechanical properties of soft rock are determined, and a contact parameter prediction model is established using machine learning algorithms (such as BPNN, SVR, XGBoost, etc.); in addition, a particle swarm optimization (PSO) algorithm is introduced to optimize the parameters of the machine learning model to improve the prediction accuracy and generalization ability of the model. This method and system have good prediction accuracy and generalization ability, providing a new idea for the contact parameter prediction method. The method mainly includes the following steps: First, a low - fidelity data set is obtained through uniaxial compression tests and Brazilian splitting tests to determine the main mesoscopic parameters affecting the macroscopic characteristics (strength and deformation) of red - bed soft rock. Secondly, based on grey relational analysis (GRA), Pearson correlation analysis, and mutual information (MI), the key contact parameters affecting the macroscopic characteristics of soft rock are determined, and a high - fidelity data set is obtained. Finally, a prediction model for the key contact parameters of the linear parallel - bond model is established, and the optimal prediction model is determined based on the prediction accuracy evaluation.

[0007] A first aspect of the present invention is to provide a discrete element contact parameter prediction method based on feature analysis and machine learning, wherein discrete element contact parameter prediction is performed based on an optimal prediction model for key contact parameters of a linear parallel bonding model, the method comprising:

[0008] S1, selecting a discrete element linear parallel bonding model for contact parameter calibration based on soft rock particle characteristics and mold material parameters, and determining model parameters and model parameter value ranges of the discrete element linear parallel bonding model to be calibrated;

[0009] S2, based on a small number of uniaxial compression and Brazilian splitting tests, obtain the contact parameters of the discrete element linear parallel bond model. The contact parameters are the main microscopic parameters that affect the macroscopic characteristics of the soft rock. The main microscopic parameters are used to form a low-fidelity contact parameter-soft rock macroscopic characteristic dataset. The soft rock macroscopic characteristics include strength and deformation.

[0010] S3, determine the key contact parameters affecting the macro characteristics of soft rock based on grey relational analysis GRA, Pearson correlation analysis and mutual information MI;

[0011] S4, a high-fidelity data set of contact parameters and macroscopic features is obtained based on a large number of uniaxial compression and Brazilian splitting tests;

[0012] S5, establish a prediction model for key contact parameters of discrete element linear parallel bonding model based on machine learning;

[0013] S6, determine the optimal prediction model for key contact parameters of the linear parallel bonding model.

[0014] Preferably, the S2 includes:

[0015] S21, obtaining a low-fidelity dataset of the first contact parameter and soft rock macroscopic characteristics corresponding to the discrete element linear parallel bond model based on numerical simulation of uniaxial compression test;

[0016] S22, obtain the second contact parameter-macro-feature low-fidelity data set corresponding to the discrete element linear parallel bonding model based on the numerical simulation of the Brazilian splitting test; among them, the tensile strength σ is selected t As an indicator of the Brazilian split test;

[0017] S23: splicing the first contact parameter-macro-feature low-fidelity dataset and the second contact parameter-macro-feature low-fidelity dataset to form the contact parameter-soft rock macro-feature low-fidelity dataset.

[0018] Preferably, the S21 includes:

[0019] (1) Determine the parameters required for the uniaxial compression test, including:

[0020] Geometric parameters, including: the axial length L of the specimen, the width W of the specimen, the porosity n, the radius R of the specimen, the length-to-radius ratio L / R of the specimen, the maximum size R of the particle or pore characteristic unit max and the minimum size R of the particle or pore characteristic unit max ; the maximum-to-minimum size ratio Rmax / Rmin of the particle or pore characteristic unit;

[0021] Material and deformation parameters, including: the particle density ρb, the linear effective modulus E between particles b2 , the effective bonding modulus between particles , the stiffness ratio k between particles rb2 , the effective bonding stiffness ratio between particles and the coefficient of friction μ between particles;

[0022] Strength parameters, including: the parallel bonding tensile strength σ t , the parallel bonding shear strength τ c and the internal friction angle ;

[0023] The coefficient of friction μ between the particle and the wall bw ;

[0024] (2) Determine the levels of the test factors and set the uniaxial compression test parameters under different levels of the test factors;

[0025] Determine the linear effective modulus E between particles b2 , the effective bonding modulus between particles , the stiffness ratio k between particles rb2 , the effective bonding stiffness ratio between particles , the coefficient of friction μ between particles, the parallel bonding tensile strength σ t , the parallel bonding shear strength τ c and the internal friction angle as the upper and lower limits of the variable parameters for the uniaxial compression numerical test; and different values under four levels of the test factors;

[0026] (3) Conduct a uniaxial compression test based on the orthogonal test rules and perform a numerical simulation based on the uniaxial compression test to obtain the first contact parameter corresponding to the discrete element linear parallel bond model for numerical simulation based on the uniaxial compression test - low-fidelity data of soft rock macroscopic characteristics; wherein, the uniaxial compression test includes: generating particles in a container with the geometric parameters formed by the specimen, where the particle size satisfies the Gaussian normal distribution, and setting the particle-particle and particle-wall contacts as linear model contacts to simulate the occurrence conditions of the rock; the numerical simulation based on the uniaxial compression test includes: modifying the contact parameters in the linear model according to the test parameters and setting the particle-particle contact as a linear parallel bond model contact; applying pressure to the upper and lower loading plates of the container at a constant speed v z to perform a numerical simulation of the uniaxial compression test; wherein, a triaxial test under low confining pressure is used to approximately replace the uniaxial test.

[0027] Preferably, the S3 includes:

[0028] S31. Use grey relational analysis (GRA) to determine the mesoscopic parameters that have the most influence on the macroscopic characteristics of soft rock, and preliminarily determine the key contact parameters based on grey relational analysis (GRA), Pearson correlation analysis, and mutual information (MI); including:

[0029] (1) Quantify the mesoscopic parameters by the correlation degree through grey relational analysis (GRA), including:

[0030] A. Determine the analysis sequence; the determined analysis sequence includes:

[0031] Take the macroscopic characteristic values of each uniaxial compression numerical test sample as the reference sequence of this uniaxial compression numerical test sample;

[0032] Take the contact parameter values of the discrete element linear parallel bond model of each uniaxial compression numerical test sample as the comparison sequence of this uniaxial compression numerical test sample;

[0033] B. Calculate the correlation coefficient and correlation degree of each mesoscopic parameter in the analysis sequence; including:

[0034] Based on the reference sequence and comparison sequence of all uniaxial compression numerical test samples, calculate the correlation coefficient between each contact parameter of the discrete element linear parallel bond model of each uniaxial compression numerical test sample and each macroscopic characteristic;

[0035] Based on the correlation coefficients between each contact parameter of the discrete element linear parallel bond model of all uniaxial compression numerical test samples and each macroscopic characteristic, calculate the correlation degree between each contact parameter of the discrete element linear parallel bond model and each macroscopic characteristic;

[0036] C. Take the compressive strength , Elastic modulus , Poisson's ratio Sort the influencing factor characteristics and determine the most influential mesoscopic parameters;

[0037] (2) Analyze the correlation between the most influential mesoscopic parameters and macroscopic characteristics based on Pearson correlation analysis and mutual information MI, and compare the analysis results of grey relational analysis GRA to obtain the correlation between the most influential mesoscopic parameters and macroscopic characteristics of soft rock; among them, Pearson is used to measure the linear correlation degree between two variables. The calculation formula of Pearson correlation analysis is as shown in Equation (6), and the calculation formula of mutual information (MI) is as shown in Equation (7):

[0038] (6)

[0039] In Equation (6), is the Pearson correlation coefficient between the th contact parameter of the discrete element linear parallel bond model and the th macroscopic characteristic, is the average value of the th contact parameter of the discrete element linear parallel bond model in the uniaxial compression numerical test samples, is the average value of the

[0040] th macroscopic characteristic in the

[0041] In Equation (7), is the mutual information value between the th contact parameter of the discrete element linear parallel bond model and the th macroscopic characteristic, is and the joint probability mass function of, is the th macroscopic characteristic of the discrete element linear parallel bond model all possible value ranges of, is the th contact parameter of the discrete element linear parallel bond model all possible value ranges of, represents the th macroscopic characteristic of the discrete element linear parallel bond model, represents the One contact parameter takes a value of with a probability of, that is the marginal probability mass function of denotes the th macroscopic feature of the discrete element linear parallel bond model takes a value of with a probability of, that is the marginal probability mass function of;

[0042] (3) Conduct a correlation analysis on the uniaxial compression numerical test results based on the Grey Relational Analysis (GRA), Pearson, and Mutual Information (MI);

[0043] S32. Based on the influence analysis, initially determine the key contact parameters among the preliminary key contact parameters that affect the macroscopic features of soft rock, thereby determining that the main mesoscopic parameters affecting the elastic modulus E and Poisson's ratio v are and , both of which show a linear relationship; the main mesoscopic parameters affecting the compressive strength σ f and the tensile strength σ t are the parallel bond shear strength τ c , and the parallel bond shear strength τ c can be determined by the tensile-compressive ratio k.

[0044] Preferably, the S4 includes:

[0045] S41. Obtain the first high-fidelity dataset based on a large number of uniaxial compression numerical tests;

[0046] S42. Obtain the second high-fidelity dataset based on conducting a large number of Brazilian splitting tests;

[0047] S43. Obtain the contact parameter - macroscopic feature high-fidelity dataset based on the first high-fidelity dataset and the second high-fidelity dataset.

[0048] Preferably, the S5 includes:

[0049] S51. Construct a dataset of the key contact parameter particle - particle bond effective modulus , particle - particle bond stiffness ratio , and parallel bond shear strength τ t based on the elastic modulus , compressive strength , tensile strength σ c and the tensile-compressive ratio k;

[0050] S52. Divide the contact parameter - macroscopic feature high-fidelity dataset into a training set T train and a test set T test; where the training set T train is used to establish a machine learning model and a test set T test for testing the performance of the machine learning model;

[0051] S53. Establish a machine learning model based on the training set T train including:

[0052] (1) Predict the key contact parameters, namely the effective modulus of particle-particle bonding, the stiffness ratio of particle-particle bonding, and the shear strength τ of parallel bonding based on typical machine learning prediction algorithms, c and establish a machine learning model; where the typical machine learning prediction algorithms include backpropagation neural network, support vector regression machine, and XGBoost;

[0053] (2) Optimize the hyperparameters of the discrete element linear parallel bonding model based on the particle swarm optimization algorithm to obtain the optimal hyperparameters;

[0054] (3) Input the training set T train into the machine learning model and train the machine learning model based on the optimal hyperparameters.

[0055] Preferably, the S6 includes:

[0056] S61. Calculate the prediction accuracy and error of the machine learning model;

[0057] Calculate the goodness of fit R 2 , the mean square error MSE, the root mean square error RMSE, and the mean absolute error MAE as the prediction accuracy and error indicators of the machine learning model; where:

[0058] The formula for the goodness of fit R 2 is shown in Equation (9) below:

[0059] (9)

[0060] The formula for the mean square error MSE is shown in Equation (10) below:

[0061] (10)

[0062] The formula for the root mean square error RMSE is shown in Equation (11) below:

[0063] (11)

[0064] The formula for the mean absolute error MAE is shown in Equation (12) below:

[0065] (12)

[0066] In Formulas (9) - (12), is the number of samples, is , and τ c measured values,[ is and , τ c predicted values,[ is , and τ c the average value of the measured values;

[0067] S62, evaluate the quality of the machine learning model based on the prediction accuracy and error level, so as to determine the optimal prediction model of the key contact parameters of the linear parallel bond model; where each index feature is: R 2 ranges from 0 to 1, and the closer the value is to 1, the higher the accuracy; MSE, RMSE, and MAE are all greater than 0, and the closer the value is to 0, the smaller the error.

[0068] The second aspect of the present invention provides an orbit sandwich inspection UAV vibration suppression system for implementing the method of the first aspect. The system includes:

[0069] A model and model parameter determination module (101) for selecting a discrete element linear parallel bond model for contact parameter calibration according to the soft rock particle characteristics and mold material parameter selection, and determining the model parameters and the value range of the model parameters of the discrete element linear parallel bond model to be calibrated;

[0070] A contact parameter test acquisition module (102) for obtaining the contact parameters of the discrete element linear parallel bond model based on a small number of uniaxial compression and Brazilian splitting tests. The contact parameters are the main mesoscopic parameters affecting the macroscopic characteristics of soft rock, and constitute a low-fidelity data set of contact parameters - soft rock macroscopic characteristics; where the macroscopic characteristics of soft rock include strength and deformation;

[0071] A key contact parameter determination module (103) for determining the key contact parameters affecting the macroscopic characteristics of soft rock based on grey relational analysis (GRA), Pearson correlation analysis, and mutual information (MI);

[0072] A high-fidelity data set test acquisition module (104) for obtaining a high-fidelity data set of contact parameters - macroscopic characteristics based on a large number of uniaxial compression and Brazilian splitting tests;

[0073] A machine learning model establishment module (105) for establishing a prediction model of the key contact parameters of the discrete element linear parallel bond model based on machine learning;

[0074] The optimal prediction model determination module (106) is configured to determine an optimal prediction model for the key contact parameters of the linear parallel bond model.

[0075] A third aspect of the present invention provides an electronic device, including a processor and a memory. The memory stores multiple instructions, and the processor is configured to read the instructions and execute the method as described in the first aspect.

[0076] A fourth aspect of the present invention provides a computer-readable storage medium. The computer-readable storage medium stores multiple instructions, and the multiple instructions can be read and executed by a processor to execute the method as described in the first aspect.

[0077] Advantages of the method and system of the present invention:

[0078] This method solves the problem of low efficiency in predicting contact parameters in the linear parallel bond model and solves the problem that it is difficult to select the optimal ML model relying on a single-level performance index. First, conduct a small number of uniaxial compression tests and Brazilian splitting tests to obtain a low-fidelity dataset of contact parameters - rock macroscopic characteristics of the linear parallel bond model; secondly, based on GRA, Pearson, and MI, determine the key contact parameters affecting rock macroscopic characteristics; finally, conduct a large number of tests to obtain a high-fidelity dataset, establish a prediction ML model, and determine the optimal ML model based on the accuracy evaluation of the ML model. Description of the Drawings

[0079] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the related art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the related art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0080] Figure 1 It is a flowchart of a discrete element contact parameter prediction method based on feature analysis - machine learning according to an embodiment of the present invention;

[0081] Figure 2 It is an architecture diagram of a discrete element contact parameter prediction system based on feature analysis - machine learning according to an embodiment of the present invention;

[0082] Figure 3 (a)-(c) are γ-correlation chord diagrams between contact parameters and macroscopic characteristics under three different algorithms;

[0083] Figure 4 It is an analysis diagram of the influence trend of mesoscopic parameters on the macroscopic characteristics of soft rock particles based on three types of correlation analysis algorithms;

[0084] Figure 5 (a)-(h) are dot-line graphs of the influence trends of microscopic parameters on macroscopic characteristics;

[0085] Figure 6 are the relationships of each combination of E, σf, σt, and k in the dataset with the key contact parameters;

[0086] Figure 7 (a)-(d) are the fitting results of the predicted values and measured values of the ML model on the training set;

[0087] Figure 8 (a)-(d) are the predicted scatter plots of the ML model on the test set;

[0088] Figure 9 (a)-(d) are the prediction accuracy and error evaluation results of the MAE of each ML model on the test set;

[0089] Figure 10 (a)-(d) are the prediction accuracy and error evaluation results of the RMSE of each ML model on the test set;

[0090] Figure 11 (a)-(d) are the prediction accuracy and error evaluation results of the MSE of each ML model on the test set;

[0091] Figure 12 (a)-(d) are the prediction accuracy and error evaluation results of the R2 of each ML model on the test set;

[0092] Figure 13 is the structural diagram of the electronic device provided according to the embodiment of the present invention. Detailed implementation manners

[0093] Next, the technical solutions of the present invention will be clearly and completely described in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0094] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. In addition, the terms "first", "second", and "third" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance.

[0095] In the description of the present invention, it should be noted that unless otherwise clearly specified and defined, the terms "installation", "connection", and "coupling" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, and it can be the communication inside two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0096] Embodiment 1

[0097] As Figure 1 shown, this embodiment provides a discrete element contact parameter prediction method based on feature analysis - machine learning, and predicts the discrete element contact parameters based on the optimal prediction model of the key contact parameters of the linear parallel bond model. The method includes:

[0098] S1. Select a discrete element linear parallel bond model for contact parameter calibration according to the soft rock particle characteristics and die material parameters, and determine the model parameters and the value ranges of the model parameters of the discrete element linear parallel bond model to be calibrated;

[0099] S2. Obtain the contact parameters of the discrete element linear parallel bond model based on a small number of uniaxial compression and Brazilian splitting tests. The contact parameters are the main mesoscopic parameters affecting the macroscopic characteristics of soft rock, and form a low - fidelity data set of contact parameter - soft rock macroscopic characteristics; wherein, the macroscopic characteristics of soft rock include strength and deformation.

[0100] As a preferred implementation manner, S2 includes:

[0101] S21. Obtain a first low - fidelity data set of contact parameter - soft rock macroscopic characteristics corresponding to the discrete element linear parallel bond model based on numerical simulation of uniaxial compression tests; including:

[0102] (1) Determine the parameters required for the uniaxial compression test, including:

[0103] Geometric parameters, including: the axial length L of the specimen, the width W of the specimen, the porosity n (used to describe the amount of pores inside the material, which is a parameter reflecting the mesoscopic structure characteristics of the material and has an important influence on the mechanical properties of the material), the radius R of the specimen, the length - to - radius ratio L / R of the specimen (used to characterize the aspect ratio of the specimen, and different aspect ratios may result in different mechanical responses of the material in the uniaxial compression test), the maximum size R of the particle or pore characteristic unit max and the minimum size R of the particle or pore characteristic unit max; The maximum size to minimum size ratio Rmax / Rmin of the particle or pore feature unit (used to reflect the inhomogeneity of the internal structure size of the material, which is a mesoscopic structure parameter. The larger this ratio, the greater the difference in the internal structure size of the material, which may affect the macroscopic mechanical properties such as the strength and deformation of the material);

[0104] Material and deformation parameters, including: particle density ρb, particle-particle linear effective modulus E b2 , particle-particle bonding effective modulus , particle-particle stiffness ratio k rb2 , particle-particle bonding effective stiffness ratio and particle-particle friction coefficient μ;

[0105] Strength parameters, including: parallel bonding tensile strength σ t , parallel bonding shear strength τ c and internal friction angle ;

[0106] Friction coefficient μ between particles and the wall bw : In the uniaxial compression numerical test, there is contact between particles and the wall, and μ bw also needs to be taken into account. Set μ bw to 0.1.

[0107] (2) Determine the levels of the test factors and set the uniaxial compression test parameters under different levels of the test factors;

[0108] Set the friction coefficient μ between particles and the wall bw to 0.1;

[0109] Determine that the four levels of the test factors are 1 to 4 respectively;

[0110] Determine the particle-particle linear effective modulus E b2 , particle-particle bonding effective modulus , particle-particle stiffness ratio k rb2 , particle-particle bonding effective stiffness ratio , particle-particle friction coefficient μ, parallel bonding tensile strength σ t , parallel bonding shear strength τ c and internal friction angle as the upper and lower limits of the variable parameters of the uniaxial compression numerical test; and different values under the four levels of the test factors;

[0111] (3) Conduct a uniaxial compression test based on the orthogonal test rules and perform a numerical simulation based on the uniaxial compression test to obtain the first contact parameter corresponding to the discrete element linear parallel bond model for numerical simulation based on the uniaxial compression test - low-fidelity data of soft rock macroscopic characteristics; wherein, the uniaxial compression test includes: generating particles in a container with the geometric parameters formed by the specimen, where the particle size satisfies the Gaussian normal distribution, and setting the particle-particle and particle-wall contacts as linear model contacts to simulate the occurrence conditions of the rock; the numerical simulation based on the uniaxial compression test includes: modifying the contact parameters in the linear model according to the test parameters and setting the particle-particle contact as a linear parallel bond model contact; applying pressure to the upper and lower loading plates of the container at a constant speed v z to conduct a numerical simulation based on the uniaxial compression test. In this embodiment, in order to facilitate the measurement of Poisson's ratio v, a triaxial test under low confining pressure is used to approximately replace the uniaxial test.

[0112] Elastic modulus The calculation formula is as shown in formula (1) below:

[0113] (1)

[0114] In formula (1), σ represents stress, that is, the force per unit area; represents strain, that is, the relative amount of deformation of the material;

[0115] The calculation formula of Poisson's ratio v is as shown in formula (2):

[0116] (2)

[0117] In formula (2), s represents the transverse strain, that is, the deformation of the material in the direction perpendicular to the external force; represents the axial strain, that is, the deformation of the material in the direction of the external force.

[0118] S22, obtain the second contact parameter corresponding to the discrete element linear parallel bond model for numerical simulation based on the Brazilian splitting test - low-fidelity data set of macroscopic characteristics; wherein the tensile strength σ t is selected as the index of the Brazilian splitting test.

[0119] The calculation formula of the tensile strength is as shown in formula (3):

[0120] (3)

[0121] In formula (3), F2 represents the applied tensile force, and A2 represents the tensile area.

[0122] S23. Concatenate the first contact parameter - macro - feature low - fidelity data set and the second contact parameter - macro - feature low - fidelity data set to form the contact parameter - soft - rock macro - feature low - fidelity data set.

[0123] S3. Determine the key contact parameters affecting the macro - features of soft rock based on grey relational analysis (GRA), Pearson correlation analysis, and mutual information (MI);

[0124] As a preferred implementation, the S3 includes:

[0125] S31. Use grey relational analysis (GRA) to determine the mesoscopic parameters most influential on the macro - features of soft rock, and preliminarily determine the key contact parameters from the most influential mesoscopic parameters based on grey relational analysis (GRA), Pearson correlation analysis, and mutual information (MI); including:

[0126] (1) Quantify the mesoscopic parameters by the correlation degree through grey relational analysis (GRA), including:

[0127] A. Determine the analysis sequence;

[0128] In this embodiment, the determined analysis sequence includes:

[0129] Take the macro - feature values of each uniaxial compression numerical test sample as the reference sequence of this uniaxial compression numerical test sample;

[0130] Take the contact parameter values of the discrete element linear parallel bond model of each uniaxial compression numerical test sample as the comparison sequence of this uniaxial compression numerical test sample;

[0131] B. Calculate the correlation coefficient and correlation degree of each mesoscopic parameter in the analysis sequence; including:

[0132] Based on the reference sequence and comparison sequence of all uniaxial compression numerical test samples, calculate the correlation coefficient between each contact parameter of the discrete element linear parallel bond model of each uniaxial compression numerical test sample and each macro - feature;

[0133] Based on the correlation coefficients between each contact parameter of the discrete element linear parallel bond model of all uniaxial compression numerical test samples and each macro - feature, calculate the correlation degree between each contact parameter of the discrete element linear parallel bond model and each macro - feature.

[0134] In this embodiment, assume that uniaxial compression numerical tests are carried out, and uniaxial compression numerical test samples are obtained. For the th uniaxial compression numerical test sample, the reference sequence of the th uniaxial compression numerical test sample is , respectively represent the compressive strength of the th uniaxial compression numerical test sample, elastic modulus and Poisson's ratio;

[0135] The comparison sequence of the th uniaxial compression numerical test sample is , where is the total number of contact parameters of the discrete element linear parallel bond model, is the index of the contact parameter of the discrete element linear parallel bond model, and is the value of the

[0136] th contact parameter of the discrete element linear parallel bond model of the th uniaxial compression numerical test sample; Calculate the correlation coefficient between the th contact parameter of the discrete element linear parallel bond model and the

[0137] th macroscopic feature in the

[0138] th uniaxial compression numerical test sample, and the formula is as shown in Equation (4): In Equation (4), is the correlation coefficient between the th contact parameter of the discrete element linear parallel bond model and the th macroscopic feature, is the minimum absolute difference between the th macroscopic feature and all contact parameter values in uniaxial compression numerical test samples, is the maximum absolute difference between the th macroscopic feature of the discrete element linear parallel bond model and all contact parameter values in uniaxial compression numerical test samples, is the value of the th macroscopic feature of the discrete element linear parallel bond model of the th uniaxial compression numerical test sample,

[0139] and

[0139] Based on the correlation coefficients between each contact parameter of the discrete element linear parallel bond model and each macroscopic feature for all uniaxial compression numerical test samples, calculate the correlation degree between each contact parameter of the discrete element linear parallel bond model and each macroscopic feature. The formula is as shown in Equation (5):

[0140] (5);

[0141] In Equation (5), is the correlation degree between the th contact parameter of the discrete element linear parallel bond model and the th macroscopic feature.

[0142] C. Rank the influencing factor characteristics of the compressive strength , elastic modulus , Poisson's ratio and determine the most influential mesoscopic parameters;

[0143] In this embodiment, grey relational analysis (GRA) is a multi-factor statistical analysis method used to determine the degree of association between factors. The following is its basic formula:

[0144] a. Data preprocessing (standardization or normalization)

[0145] Usually, the original data is processed by mean value method or initial value method to eliminate the dimension difference.

[0146] Mean value method: ;

[0147] Initial value method (taking the first data as the reference): ;

[0148] b. Calculate the correlation coefficient

[0149] Suppose the reference sequence is , and the comparison sequence is . Then the correlation coefficient formula is: ;

[0150] Where: : The absolute difference between the reference sequence and the comparison sequence at the kth point; : The global minimum difference; : The global maximum difference; is the resolution coefficient (usually taken as 0.5, range (0,1)).

[0151] c. Calculate the correlation degree

[0152] The correlation degree is the average value of the correlation coefficients, reflecting the overall degree of association:

[0153] ;

[0154] d. Relevance ranking

[0155] Sort by relevance Sort from largest to smallest. The larger the value, the stronger the correlation between the comparison sequence and the reference sequence.

[0156] (2) Analyze the correlation between the most influential mesoscopic parameters and macroscopic characteristics based on Pearson correlation analysis and mutual information (MI), and compare the results of GRA analysis to more accurately obtain the correlation between the most influential mesoscopic parameters and macroscopic characteristics of soft rock. Among them, Pearson is used to measure the degree of linear correlation between two variables. The calculation formula of Pearson correlation analysis is shown in Equation (6), and the calculation formula of mutual information (MI) is shown in Equation (7).

[0157] The calculation process of Pearson correlation analysis is shown in Equation (6):

[0158] (6)

[0159] In Equation (6), is the Pearson correlation coefficient between the th contact parameter of the discrete element linear parallel bond model and the th macroscopic characteristic, is the average value of the th contact parameter of the discrete element linear parallel bond model in uniaxial compression numerical test samples, is the average value of the

[0160] th macroscopic characteristic in

[0161] (7);

[0162] In Equation (7), is the mutual information value between the th contact parameter of the discrete element linear parallel bond model and the th macroscopic characteristic, is the and joint probability mass function, is the th macroscopic characteristic all possible value ranges, is the th contact parameter all possible value ranges, Represents the th macroscopic feature of the discrete element linear parallel bond model, Represents the th contact parameter of the discrete element linear parallel bond model, Represents the th contact parameter Taking the value of The probability of, that is, The marginal probability mass function of, Represents the th macroscopic feature of the discrete element linear parallel bond model Taking the value of The probability of, that is, The marginal probability mass function of.

[0163] Mutual Information (MI) is an index in information theory to measure the statistical dependence between two random variables, reflecting the amount of information contained in one variable about another. The following are its core formula and explanations:

[0164] Definition of mutual information

[0165] The formula for mutual information I(X;Y) is: ;

[0166] In the formula: p(x,y) represents the joint probability distribution of X and Y; p(x) and p(y) represent the marginal probability distributions of X and Y; the logarithm base is usually taken as 2 (unit: bit) or the natural logarithm e (unit: nat).

[0167] (3) Conduct a correlation analysis on the uniaxial compression numerical test results based on GRA, Pearson, and MI.

[0168] S32, Based on the influence analysis, initially determine the key contact parameters that affect the macroscopic features of soft rock among the key contact parameters, so as to determine that the main mesoscopic parameters affecting the elastic modulus E and Poisson's ratio v are and , both of which show a linear relationship; while the main mesoscopic parameters affecting the compressive strength σ f、 Tensile strength σ t The main mesoscopic parameter of is the parallel bond shear strength τ c , the parallel bond shear strength τ c Can be determined by the tensile-compressive ratio k.

[0169] The calculation formula for the compressive strength σ f Is shown in Equation (8):

[0170] (8);

[0171] In formula (8), F1 represents the applied force, and A1 represents the compression area.

[0172] In this embodiment, S32 includes: analyzing the influence trend of mesoscopic parameters on the macroscopic characteristics of soft rock particles based on three types of correlation analysis algorithms, obtaining the main mesoscopic parameters affecting the macroscopic characteristics, and further determining the key contact parameters affecting the macroscopic characteristics.

[0173] S4. Obtain a high-fidelity dataset of contact parameters - macroscopic characteristics based on a large number of uniaxial compression and Brazilian splitting tests;

[0174] As a preferred embodiment, S4 includes:

[0175] S41. Obtain a first high-fidelity dataset based on a large number of uniaxial compression numerical tests;

[0176] S42. Obtain a second high-fidelity dataset based on conducting a large number of Brazilian splitting tests;

[0177] S43. Obtain a high-fidelity dataset of contact parameters - macroscopic characteristics based on the first high-fidelity dataset and the second high-fidelity dataset.

[0178] S5. Establish a prediction model for the key contact parameters of the discrete element linear parallel bond model based on machine learning;

[0179] As a preferred embodiment, S5 includes:

[0180] S51. Based on the elastic modulus , compressive strength , tensile strength σ t and the compression-tension ratio k, construct a dataset of the key contact parameter particle-particle bond effective modulus , particle-particle bond stiffness ratio and the parallel bond shear strength τ c ;

[0181] In this embodiment, by analyzing the relationship between different macroscopic characteristics of soft rock and the key contact parameters, establish a dataset D={(x,y)} of the key contact parameter particle-particle bond effective modulus , particle-particle bond stiffness ratio and the parallel bond shear strength τ c , where x is the input feature of the prediction model, composed of the elastic modulus , compressive strength , tensile strength σ t and the compression-tension ratio k, and the output feature is composed of the key contact parameter particle-particle bond effective modulus , particle-particle bond stiffness ratio and the parallel bond shear strength τc Composition;

[0182] S52, divide the contact parameter - macroscopic feature high - fidelity data set into a training set T train and a test set T test ; wherein, the training set T train is used to establish a machine learning model and the test set T test is used to test the performance of the machine learning model.

[0183] In this embodiment, the contact parameter - macroscopic feature high - fidelity data set is divided into a training set and a test set according to a ratio of 7:3, where the training set is used to establish a machine learning model, and the test set is used to test the performance of the machine learning model;

[0184] S53, establish a machine learning model based on the training set T train , including:

[0185] (1) Predict the effective modulus of particle - particle bonding, the stiffness ratio of particle - particle bonding and the shear strength τ of parallel bonds c of key contact parameters based on typical machine learning prediction algorithms, and establish a machine learning model; wherein, the typical machine learning prediction algorithms include Back Propagation Neural Network (BPNN), Support Vector Machine (SVM) and XGBoost;

[0186] (2) Optimize the hyperparameters of the discrete element linear parallel bond model based on the Particle Swarm Optimization (PSO) algorithm to obtain the optimal hyperparameters;

[0187] (3) Input the training set T train into the machine learning model, and train the machine learning model based on the optimal hyperparameters.

[0188] S6, determine the optimal prediction model of the key contact parameters of the linear parallel bond model.

[0189] Since the prediction results on the training set only represent the prediction ability during the modeling process of the machine learning model and cannot yet reflect the generalization ability of the model, the prediction performance of the machine learning model should be judged through the test set.

[0190] As a preferred implementation, the S6 includes:

[0191] S61, calculate the prediction accuracy and error of the machine learning model;

[0192] Calculate the goodness of fit R 2 , the mean square error MSE, the root mean square error RMSE, and the mean absolute error MAE as the prediction accuracy and error metrics of the machine learning model; where:

[0193] The goodness of fit R 2 The formula is as shown in Equation (9) below:

[0194] (9)

[0195] The formula for the mean square error MSE is as shown in Equation (10) below:

[0196] (10)

[0197] The formula for the root mean square error RMSE is as shown in Equation (11) below:

[0198] (11)

[0199] The formula for the mean absolute error MAE is as shown in Equation (12) below:

[0200] (12)

[0201] In Equations (9)-(12), is the number of samples, is 、 and τ c measured values, is and 、τ c predicted values, is 、 and τ c the average value of the measured values;

[0202] S62, evaluate the quality of the machine learning model based on the prediction accuracy and error level, so as to determine the optimal prediction model of the key contact parameters of the linear parallel bond model; where the characteristics of each index are that R 2 ranges from 0 to 1, and the closer the value is to 1, the higher the accuracy; MSE, RMSE, and MAE are all greater than 0, and the closer the value is to 0, the smaller the error.

[0203] Example 2

[0204] As Figure 2 shown, this embodiment provides an orbit sandwich inspection UAV vibration suppression system for implementing the method of Embodiment 1. The system includes:

[0205] The model and model parameter determination module 101 is used to select a discrete element linear parallel bond model for contact parameter calibration according to the soft rock particle characteristics and die material parameters, and determine the model parameters and the value ranges of the model parameters of the discrete element linear parallel bond model to be calibrated;

[0206] The contact parameter test acquisition module 102 is used to obtain the contact parameters of the discrete element linear parallel bond model based on a small number of uniaxial compression and Brazilian splitting tests. The contact parameters are the main mesoscopic parameters affecting the macroscopic characteristics of soft rock, and the main mesoscopic parameters are used to form a low-fidelity data set of contact parameter-soft rock macroscopic characteristics; among them, the macroscopic characteristics of soft rock include strength and deformation;

[0207] The key contact parameter determination module 103 is used to determine the key contact parameters affecting the macroscopic characteristics of soft rock based on grey relational analysis (GRA), Pearson correlation analysis, and mutual information (MI);

[0208] The high-fidelity data set test acquisition module 104 is used to obtain a high-fidelity data set of contact parameter-macroscopic characteristics based on a large number of uniaxial compression and Brazilian splitting tests;

[0209] The machine learning model establishment module 105 is used to establish a prediction model for the key contact parameters of the discrete element linear parallel bond model based on machine learning;

[0210] The optimal prediction model determination module 106 is used to determine the optimal prediction model for the key contact parameters of the linear parallel bond model.

[0211] Application example

[0212] This embodiment discloses a discrete element contact parameter prediction method based on machine learning, including the following steps: First, a low-fidelity data set is obtained through uniaxial compression tests and Brazilian splitting tests to determine the main mesoscopic parameters affecting the macroscopic characteristics (strength and deformation) of red-bed soft rock. Secondly, based on GRA, Pearson, and MI, the key contact parameters affecting the macroscopic characteristics of soft rock are determined, and a high-fidelity data set is obtained. Finally, a prediction model for the key contact parameters of the linear parallel bond model is established, and the optimal prediction model is determined based on the prediction accuracy evaluation.

[0213] The embodiments and implementation processes implemented according to the complete method content of the present invention are as follows:

[0214] 1) Select discrete element linear parallel bond according to the soft rock particle characteristics and die material parameters. Determine the parameters and value ranges of the discrete element contact model to be calibrated.

[0215] 2) Uniaxial compression numerical test

[0216] In the uniaxial compression numerical test, there is contact between particles and the wall, and μ needs to be...bw is also taken into account, and μ bw is set to 0.1, and the remaining 8 parameters (the linear effective modulus E of particle-particle b2 , the bonding effective modulus of particle-particle , the stiffness ratio k of particle-particle rb2 , the bonding effective stiffness ratio of particle-particle , the friction coefficient μ of particle-particle, the tensile strength σ of parallel bond t , the shear strength τ of parallel bond c , the internal friction angle ) are used as variable parameters for the uniaxial compression numerical test, and the upper and lower limits of the 8 parameters are determined. The values of the four levels of each test factor are determined as shown in Table 1. Using the orthogonal test design, select the L 96 4 8 (8 factors, 4 levels, 96 tests) type orthogonal table to carry out the uniaxial compression numerical test of red-bed soft rock.

[0217] Table 1

[0218]

[0219] Table 2

[0220]

[0221] The process of uniaxial compression modeling is as follows: Generate particles in a cylindrical container with L = 100 mm and W = 50 mm, where the particle size satisfies the Gaussian normal distribution, and R max / R min is 1.66, and R is 0.8 mm, and the particle-particle and particle-wall contacts are set as linear model contacts to simulate the occurrence conditions of the rock. Modify the contact parameters in the linear model according to the test parameters, and set the particle-particle contact as a linear parallel bond model contact. Apply a constant velocity v z on the upper and lower loading plates. To ensure the stability of the model, v z needs to be controlled at a small value.

[0222] 3) Use correlation analysis for the mesoscopic parameters that have the most influence on the macroscopic characteristics of the rock.

[0223] Based on GRA, Pearson, and MI, perform a correlation degree analysis on the results of 96 groups of uniaxial compression numerical tests of red-bed soft rock. There are individual mesoscopic parameters that have a greater influence on the macroscopic characteristics of compressive strength σ f , elastic modulus E, and Poisson's ratio v. And the maximum correlation degrees between the mesoscopic parameters and σ f and E are both greater than 0.7, while the maximum correlation degree between it and v is 0.56, indicating that in the material strength and deformation characteristics, the mesoscopic parameters mainly affect the material E and σ fIn addition, there is a certain gap between the correlation results of different algorithms, so the average value of the results is taken to obtain the microscopic parameters and σ f , the correlation between E and v.

[0224] The γ correlation between contact parameters and macroscopic characteristics under four different algorithms is as follows: Figure 3 (a)-(c) are shown. Sorting the averaged γ from large to small shows that the microscopic parameters that have a greater impact on E are 、k rb2 and ; for σ f The microscopic parameters with greater influence are μ and σ t , τ c and ; The microscopic parameters that have a greater impact on v are 、E b2 、k rb2 and Among them, the parameters that have a greater impact on E also affect v.

[0225] In order to further explore the relationship between the microscopic parameters and σ f , E and v, and the relationship between them is studied by controlling the variable method. When a certain microscopic parameter with a greater influence is changed, the other parameters take the minimum value (factor level 1), the middle value (factor levels 2 and 3) or the maximum value (factor level 4). The results are as follows Figure 5 shown. Figure 5 (a)~(h) are the relationships between the microstructure and E, which have a greater impact on E. It is linearly related to E with a slope of about 1.1, while the intercept is controlled by the other parameters; rb2 and The increase shows a slight downward trend, indicating that k rb2 and The effect on E is not significant.

[0226] Based on the three types of correlation analysis algorithms and the trend analysis of the influence of microscopic parameters on the macroscopic characteristics of red bed soft rocks, such as Figure 4 As shown, the key microscopic parameters affecting the macroscopic characteristics E are , affecting the compressive strength σ f and tensile strength σ t The key microscopic parameter is the bond shear strength τ c , the key microscopic parameters affecting v are and k rb2 In addition, due to and k rb2 The influence trend on v is consistent, so we can =k rb2 .

[0227] In summary, through the analysis of the correlation mechanism between a series of microscopic parameters and macroscopic characteristics, the main microscopic parameters affecting the elastic modulus E and Poisson's ratio v are and , both of which show a linear relationship; while the main microscopic parameters affecting the compressive strength σ f、 and the tensile strength σ t are τ c , and τ c can be determined by k.

[0228] 4) Use typical ML algorithms to predict E, σ f , σ t . Introduce the PSO algorithm to optimize the hyperparameters of the model, and train the ML model based on the optimal hyperparameters.

[0229] By analyzing the data of soft rocks with different microscopic parameters and key contact parameters, establish , and τ c datasets. The relationships between various key contact parameter combinations in the dataset for , and τ c are as shown in Figure 6 .

[0230] Divide the dataset into a training set T train and a test set T test in a ratio of 7:3.

[0231] The fitting results of the predicted values and measured values of each ML model on the training set are as shown in Figure 7 (a)-(d). The fitting effects of the three ML models on the measured values of , and τ c on the training set are all good. Basically, they fluctuate around the 45° central axis, and most points are concentrated within the 10% error line, with only a few measurement points outside the error line, indicating that the fitting errors of each ML model are relatively low. From Figure 8 (a)-(d), it can be seen that among the three ML models, the one with the highest training accuracy is XGBoost.

[0232] 5) Evaluate each ML model by combining the prediction accuracy and error of the ML model to obtain the optimal ML model.

[0233] The predicted scatter plots of each ML model on the test set are as shown in Figure 9 (a)-(d). The prediction effects of the three ML models on the test set are all good, and they basically fluctuate around the 45° central axis. Among them, the one most concentrated on the central axis is the XGBoost model, followed by the BPNN model, and finally the SVR model.

[0234] The prediction accuracy and error evaluation results of each ML model on the test set are as follows Figure 10 in (a)-(d)- Figure 12 as shown in (a)-(d). On the training set, the MSE, RMSE, and MAE of each ML model are relatively small. Among them, the R values of each ML model for elastic modulus and k 2 are all greater than 0.9, demonstrating a relatively high fitting accuracy. From Figure 10 in (a)-(d)- Figure 12 in (a)-(d), it can be seen that the MAE, RMSE, and MSE of the BPNN and XGBoost models for elastic modulus are relatively small. The XGBoost model for compressive strength has the smallest MAE, the BPNN model has the smallest RMSE and MAE, and the SVR model has relatively large MAE, RMSE, and MSE. For tensile strength, the XGBoost model has the smallest MAE, RMSE, and MSE, followed by the BPNN model, and finally the SVR model. Therefore, it can be concluded from the analysis that among the three ML models, XGBoost has the highest test accuracy.

[0235] Based on the prediction accuracy and error evaluation of the comprehensive ML model, it can be seen that the XGBoost model is the best. Therefore, the XGBoost model is selected as the optimal ML model.

[0236] The present invention also provides a memory storing multiple instructions for implementing the method of Embodiment 1.

[0237] As Figure 13 shown, the present invention also provides an electronic device including a processor 301 and a memory 302 connected to the processor 301. The memory 302 stores multiple instructions that can be loaded and executed by the processor, enabling the processor to execute the method of Embodiment 1.

[0238] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A discrete element contact parameter prediction method based on feature analysis - machine learning, characterized in that Predicting discrete element contact parameters based on an optimal prediction model for key contact parameters of a linear parallel bond model, the method comprising: S1. Selecting a discrete element linear parallel bond model for contact parameter calibration according to the characteristics of soft rock particles and die material parameters, and determining the model parameters and the value ranges of the model parameters of the discrete element linear parallel bond model to be calibrated; S2. Obtaining the contact parameters of the discrete element linear parallel bond model based on a small number of uniaxial compression and Brazilian splitting tests, where the contact parameters are the main mesoscopic parameters affecting the macroscopic characteristics of soft rock, and forming a low-fidelity data set of contact parameters - macroscopic characteristics of soft rock with the main mesoscopic parameters; wherein, the macroscopic characteristics of soft rock include strength and deformation; S3. Determining the key contact parameters affecting the macroscopic characteristics of soft rock based on grey relational analysis (GRA), Pearson correlation analysis, and mutual information (MI); S4. Obtaining a high-fidelity data set of contact parameters - macroscopic characteristics based on a large number of uniaxial compression and Brazilian splitting tests; S5. Establishing a prediction model for key contact parameters of a discrete element linear parallel bond model based on machine learning; S6. Determining an optimal prediction model for key contact parameters of the linear parallel bond model.

2. The discrete element contact parameter prediction method based on feature analysis - machine learning according to claim 1, characterized in that The S2 includes: S21. Obtaining a first low-fidelity data set of contact parameters - macroscopic characteristics of soft rock corresponding to the discrete element linear parallel bond model obtained by numerical simulation based on uniaxial compression tests; S22, obtain a second contact parameter-macro feature low-fidelity data set corresponding to the discrete element linear parallel bond model based on numerical simulation of the Brazilian splitting test; where the tensile strength σ t is selected as the Brazilian splitting test index; S23. Stitching together the first low-fidelity data set of contact parameters - macroscopic characteristics and the second low-fidelity data set of contact parameters - macroscopic characteristics to form the low-fidelity data set of contact parameters - macroscopic characteristics of soft rock.

3. The discrete element contact parameter prediction method based on feature analysis - machine learning according to claim 2, characterized in that The S21 includes: (1) Determining the parameters required for uniaxial compression tests, including: Geometric parameters, including: the axial length L of the specimen, the width W of the specimen, the porosity n, the radius R of the specimen, the length-radius ratio L / R of the specimen, the maximum size R of the particle or pore characteristic unit max and the minimum size R of the particle or pore characteristic unit max ; the maximum size-minimum size ratio Rmax / Rmin of the particle or pore characteristic unit; Material and deformation parameters, including: particle density ρb, particle-particle linear effective modulus E b2 , particle-particle bonding effective modulus , particle-particle stiffness ratio k rb2 , particle-particle bonding effective stiffness ratio and particle-particle friction coefficient μ; Strength parameters, including: parallel bond tensile strength σ t , parallel bond shear strength τ c and internal friction angle ; Coefficient of friction μ between the particle and the wall bw ; (2) Determining the levels of test factors and setting the uniaxial compression test parameters at different levels of test factors; Determine the linear effective modulus E of particle-particle b2 , the effective bonding modulus of particle-particle , the stiffness ratio k of particle-particle rb2 , the effective bonding stiffness ratio of particle-particle , the friction coefficient μ of particle-particle, the tensile strength σ of parallel bonding t , the shear strength τ of parallel bonding c and the internal friction angle as the upper and lower limits of the variable parameters in the uniaxial compression numerical test; and different numerical values at four test factor levels; (3) Conduct uniaxial compression tests based on orthogonal test rules, and perform numerical simulations based on the uniaxial compression tests to obtain the first contact parameters corresponding to the discrete element linear parallel bond model for numerical simulations based on the uniaxial compression tests - low-fidelity data of the macroscopic characteristics of soft rock; wherein, the uniaxial compression test includes: generating particles in a container with the geometric parameters formed by the specimen, where the particle size satisfies a Gaussian normal distribution, and setting the particle-particle and particle-wall contacts as linear model contacts to simulate the occurrence conditions of the rock; the numerical simulation based on the uniaxial compression test includes: modifying the contact parameters in the linear model according to the test parameters, and setting the particle-particle contact as a linear parallel bond model contact; applying pressure to the upper and lower loading plates of the container at a constant speed v z to conduct numerical simulations of the uniaxial compression test; wherein, a triaxial test under low confining pressure is used to approximately replace the uniaxial test.

4. The discrete element contact parameter prediction method based on feature analysis - machine learning according to claim 3, characterized in that The S3 includes: S31. Using grey relational analysis (GRA) to determine the mesoscopic parameters most influential on the macroscopic characteristics of soft rock, and preliminarily determining the key contact parameters from the most influential mesoscopic parameters based on grey relational analysis (GRA), Pearson correlation analysis, and mutual information (MI); including: (1) Quantifying the mesoscopic parameters by correlation degree through grey relational analysis (GRA), including: A. Determining the analysis sequences; the determined analysis sequences include: Taking the macroscopic characteristic values of each uniaxial compression numerical test sample as the reference sequence of the uniaxial compression numerical test sample; Taking the contact parameter values of the discrete element linear parallel bond model of each uniaxial compression numerical test sample as the comparison sequence of the uniaxial compression numerical test sample; B. Calculating the correlation coefficients and correlation degrees of each mesoscopic parameter in the analysis sequences; including: Based on the reference sequences and comparison sequences of all uniaxial compression numerical test samples, calculating the correlation coefficients between each contact parameter of the discrete element linear parallel bond model of each uniaxial compression numerical test sample and each macroscopic characteristic; Based on the correlation coefficients between each contact parameter of the discrete element linear parallel bond model and each macroscopic characteristic of all uniaxial compression numerical test samples, calculating the correlation degrees between each contact parameter of the discrete element linear parallel bond model and each macroscopic characteristic; C. Rank the influencing factor characteristics of the compressive strength , elastic modulus , Poisson's ratio and determine the most influential mesoscopic parameters; (2) Analyze the correlation between the most influential mesoscopic parameters and macroscopic characteristics based on Pearson correlation analysis and mutual information MI, and compare the analysis results of grey relational analysis GRA to obtain the correlation between the most influential mesoscopic parameters and macroscopic characteristics of soft rock; among them, Pearson is used to measure the linear correlation degree between two variables. The calculation formula of Pearson correlation analysis is as shown in formula (6), and the calculation formula of mutual information (MI) is as shown in formula (7): （6）； In Equation (6), is the Pearson correlation coefficient between the -th contact parameter and the -th macroscopic feature of the discrete element linear parallel bond model, is the average value of the -th contact parameter of the discrete element linear parallel bond model in numerical uniaxial compression test samples, is the average value of the -th macroscopic feature in （7）； In Equation (7), is the mutual information value between the th contact parameter and the th macroscopic feature of the discrete element linear parallel bond model, is and 's joint probability mass function, is the range of all possible values of the th macroscopic feature of the discrete element linear parallel bond model, is the range of all possible values of the th contact parameter of the discrete element linear parallel bond model, represents the th macroscopic feature of the discrete element linear parallel bond model, represents the th contact parameter of the discrete element linear parallel bond model, represents the th contact parameter of the discrete element linear parallel bond model taking the value of , that is, 's marginal probability mass function, represents the th macroscopic feature of the discrete element linear parallel bond model taking the value of , that is, 's marginal probability mass function; (3) Conduct a correlation degree analysis on the uniaxial compression numerical test results based on grey relational analysis GRA, Pearson, and mutual information MI; S32. Based on the influence analysis, initially determine the key contact parameters that affect the macroscopic characteristics of soft rock among the initially determined key contact parameters, so as to determine that the main mesoscopic parameters affecting the elastic modulus E and Poisson's ratio v are and , and both show a linear relationship; the main mesoscopic parameters affecting the compressive strength σ f and the tensile strength σ t are the parallel bond shear strength τ c , and the parallel bond shear strength τ c can be determined by the tensile-compressive ratio k.

5. A discrete element contact parameter prediction method based on feature analysis - machine learning according to claim 4, characterized in that The S4 includes: S41, obtaining the first high-fidelity dataset based on a large number of uniaxial compression numerical tests; S42, obtaining the second high-fidelity dataset based on conducting a large number of Brazilian splitting tests; S43, obtaining the contact parameter-macroscopic characteristic high-fidelity dataset based on the first high-fidelity dataset and the second high-fidelity dataset.

6. The discrete element contact parameter prediction method based on feature analysis - machine learning according to claim 5, wherein, The S5 includes: S51, based on the elastic modulus , compressive strength , tensile strength σ t and the compression-tension ratio k to construct a dataset of key contact parameters, including the effective modulus of particle-particle bonding , the stiffness ratio of particle-particle bonding and the shear strength τ of parallel bonds c ; S52, divide the contact parameter - macroscopic feature high - fidelity data set into a training set T train and a test set T test ; among them, the training set T train is used to establish a machine learning model, and the test set T test is used to test the performance of the machine learning model; S53. Establish a machine learning model based on the training set T train as follows: (1) Predict the key contact parameters, namely the effective modulus of particle-particle bonding , the particle-particle bonding stiffness ratio , and the parallel bond shear strength τ c based on typical machine learning prediction algorithms, and establish a machine learning model. Among them, the typical machine learning prediction algorithms include backpropagation neural network, support vector regression machine, and XGBoost;​​​​​​ (2) Optimize the hyperparameters of the discrete element linear parallel bond model based on the particle swarm optimization algorithm to obtain the optimal hyperparameters; (3) Input the training set T train into the machine learning model, and train the machine learning model based on the optimal hyperparameters.

7. A discrete element contact parameter prediction method based on feature analysis - machine learning according to claim 6, characterized in that The S6 includes: S61, calculating the prediction accuracy and error of the machine learning model; Calculate the goodness of fit R 2 , the mean squared error MSE, the root mean squared error RMSE, and the mean absolute error MAE are used as the prediction accuracy and error metrics of the machine learning model; where: The goodness of fit R 2 The formula is as shown in the following formula (9): （9） The formula for the mean square error MSE is as shown in formula (10) below: (10) The formula for the root mean square error RMSE is as shown in formula (11) below: (11) The formula for the mean absolute error MAE is as shown in formula (12) below: (12) In Equations (9) to (12), is the number of samples, is , and the measured values of τ c ; is the predicted value of Ē b and k̄ rb , τ c ; is , and the average value of the measured values of τ c . S62. Evaluate the quality of the machine learning model based on the prediction accuracy and error level, so as to determine the optimal prediction model for the key contact parameters of the linear parallel bond model; where each index feature is: R 2 The range is between 0 and 1, and the closer the value is to 1, the higher the accuracy; MSE, RMSE, and MAE are all greater than 0, and the closer the value is to 0, the smaller the error.

8. An orbit mezzanine inspection UAV vibration suppression system for implementing the method according to any one of claims 1-7, characterized in that, The system includes: A model and model parameter determination module (101), which is used to select a discrete element linear parallel bond model for contact parameter calibration according to the soft rock particle characteristics and die material parameter selection, and determine the model parameters and the value range of the model parameters of the discrete element linear parallel bond model to be calibrated; A contact parameter test acquisition module (102), which is used to obtain the contact parameters of the discrete element linear parallel bond model based on a small number of uniaxial compression and Brazilian splitting tests. The contact parameters are the main mesoscopic parameters affecting the macroscopic characteristics of soft rock, and form a contact parameter-soft rock macroscopic characteristic low-fidelity dataset with the main mesoscopic parameters; among them, the macroscopic characteristics of soft rock include strength and deformation; A key contact parameter determination module (103), which is used to determine the key contact parameters affecting the macroscopic characteristics of soft rock based on grey relational analysis (GRA), Pearson correlation analysis, and mutual information (MI); A high-fidelity dataset test acquisition module (104), which is used to obtain a contact parameter-macroscopic characteristic high-fidelity dataset based on a large number of uniaxial compression and Brazilian splitting tests; A machine learning model establishment module (105), which is used to establish a prediction model for the key contact parameters of the discrete element linear parallel bond model based on machine learning; An optimal prediction model determination module (106), which is used to determine the optimal prediction model for the key contact parameters of the linear parallel bond model.

9. An electronic device, characterized in that, It includes a processor and a memory. The memory stores multiple instructions, and the processor is used to read the instructions and execute the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores multiple instructions, and the multiple instructions can be read and executed by the processor to execute the method according to any one of claims 1-7.

Images